Full, Quick, and Duo Deliberation Modes in the Council of High Intelligence: Key Distinctions

Full mode orchestrates 3–18 members through three rounds of blind cross-examination for high-stakes decisions, Quick mode accelerates 2–11 members through two rapid rounds without anonymization, and Duo mode forces exactly two members into a direct dialectic to surface opposing viewpoints.

The Council of High Intelligence is an open-source multi-agent deliberation framework that routes queries across diverse LLM providers to simulate expert consensus. Its full, quick, and duo deliberation modes provide distinct architectural pathways for different decision complexities, each implemented with specific round structures, voting logic, and synthesis protocols in SKILL.md and README.md.

Architectural Overview of the Three Modes

Invocation Flags and Panel Composition

Each mode is triggered by a specific command-line flag that determines the panel size and selection logic:

• Full Mode (--full or default): Accepts 3–18 members. When invoked without constraints, it defaults to all 18 council members, though users can limit the panel using --triad, --members, or a specific profile. This mode is defined in SKILL.md at lines 61–66 as the default coordination pipeline.

• Quick Mode (--quick): Restricts the panel to 2–11 members, auto-selecting a triad unless a specific panel is passed. Defined at line 31 of both README.md and SKILL.md, this mode collapses the problem-restate gate into the initial prompt to reduce latency.

• Duo Mode (--duo): Forces exactly two members, chosen either by the --members flag or by algorithmically matching a polarity-pair to the problem. Implemented at line 6 of SKILL.md, this mode eliminates anonymization because the dialectic relies on explicit tension between identified opponents.

Round Structure and Cross-Examination

The depth of deliberation varies significantly across the three protocols:

Full Mode executes a 7-step sequence across three rounds:

1. Independent analysis (blind-first)
2. Cross-examination with identity masking and an "anti-conformity" directive forcing members to engage at least two peers
3. Final crystallization with weighted voting

Quick Mode compresses this into a 4-step sequence with only two rounds:

1. Rapid analysis combining problem restatement and core analysis
2. Final positions delivered anonymously, but without the structured cross-examination phase found in full mode

Duo Mode returns to three rounds but with a dialectic structure:

1. Opening positions (problem restate + position)
2. Direct response where each member replies to the other's opening
3. Final statements synthesizing the dialectic

In SKILL.md, the full mode sequence spans steps 0–7, while quick mode uses steps 0–3 and duo mode uses steps 0–4.

Verdict Synthesis and Voting Mechanisms

Each mode produces a distinct verdict format through a Chairman selected by the routing rules defined at STEP 1.7 in SKILL.md:

• Full Mode Verdict: The Chairman synthesizes a comprehensive document including vote tallies, unresolved questions, and kill criteria. Weighted voting applies structured STANCE: lines at 1.0 per member, with a 1.5× domain-weight multiplier applied before final tallying at STEP 6.

• Quick Mode Verdict: Produces a condensed "Quick Verdict" containing only a concrete recommendation and kill criteria. While the same weighted tally logic applies, the smaller panel size means the domain-weight seat often unilaterally decides the outcome.

• Duo Mode Verdict: Generates a "Duo Verdict" presenting both positions side-by-side without weighted tallying. The user retains final authority to select which opposing perspective to follow, as the protocol explicitly avoids synthesizing a false consensus from two fundamentally opposed views.

Problem-Restate Gate Implementation

The handling of the mandatory problem-restate step reveals architectural intent:

• Full Mode: Implements a separate Step 1.5 where every member must restate the problem before any analysis begins, ensuring cognitive diversity in problem framing.
• Quick Mode: Embeds the restatement implicitly within the Rapid-Analysis prompt to minimize token usage and latency.
• Duo Mode: Integrates the restate into the Opening-Positions prompt, as shown in the duo-mode snippet from SKILL.md:

You are operating as one half of a structured dialectic with one opponent.
Read your agent definition at ~/.claude/agents/council-{name}.md and follow it precisely.

The problem under deliberation:
{problem}

First, in ONE sentence, restate this problem through your analytical lens. Then state your position using your Output Format (Standalone).
Limit: 300 words maximum.

Source Code Implementation

All three modes share the provider auto-routing logic described in the README's "Multi-Provider Auto-Routing" section and the Chairman selection algorithm at STEP 1.7, but diverge in their execution pipelines:

• Full mode follows the 7-step coordination pipeline: provider routing → problem restate → Round 1 → Round 2 (cross-examination with enforcement) → Round 3 → verdict synthesis.

• Quick mode begins at line 31 of SKILL.md, collapsing the restate gate into Round 1 and skipping cross-examination entirely to reach verdict in 3 steps.

• Duo mode starts at line 6 of SKILL.md, forcing a two-member constraint check before executing the dialectic prompts without anonymization layers.

Practical Usage Examples

The repository provides specific invocation patterns for each mode in the README's Example Usage section (lines 23–49):

# Full mode – default (uses auto-selected triad or full panel)

/council --triad strategy "Should we launch a premium tier?"

# Quick mode – one-shot rapid analysis

/council --quick "Add Redis caching to the auth flow?"

# Duo mode – explicit polarity pair dialectic

/council --duo "Rewrite the monolith as microservices?"

The quick-mode prompt structure in SKILL.md enforces brevity:

You are operating as a council member in a rapid deliberation.
Read your agent definition at ~/.claude/agents/council-{name}.md and follow it precisely.

The problem under deliberation:
{problem}

First, in ONE sentence, restate this problem through your analytical lens. Then produce a condensed analysis:
- Essential Question (1-2 sentences)
- Your core analysis (key insight only)
- Verdict (direct recommendation)
- Confidence (High/Medium/Low)

Limit: 200 words maximum. Be decisive.

Summary

• Full mode provides maximum deliberative depth with 3–18 members, three rounds including anonymized cross-examination, and comprehensive verdict synthesis with weighted voting.
• Quick mode prioritizes speed with 2–11 members, two rounds without cross-examination, and condensed verdicts suitable for lower-complexity decisions.
• Duo mode facilitates structured dialectic between exactly two members across three rounds, eliminating anonymization and weighted voting to preserve opposing viewpoints.
• All modes utilize the same provider auto-routing logic and Chairman selection algorithm defined in SKILL.md and README.md.

Frequently Asked Questions

When should I use full mode versus quick mode?

Use full mode for high-stakes architectural or product strategy decisions requiring exhaustive challenge and a rich audit trail of dissent. Use quick mode for rapid, lower-complexity questions where speed outweighs the need for multi-round cross-examination, such as tactical implementation details or straightforward feature additions.

How does voting work in duo mode if there are only two members?

Duo mode does not use weighted voting. The two members present opposing positions side-by-side in the final verdict, and the user manually selects which perspective to follow. This design preserves the dialectic tension rather than forcing an artificial consensus from two fundamentally opposed views.

Can I customize the panel size in each deliberation mode?

Yes, within the constraints of each mode. Full mode accepts 3–18 members and can be limited using --triad or --members. Quick mode accepts 2–11 members. Duo mode strictly requires exactly two members, specified via --members or auto-selected based on polarity matching to the problem domain.

What is the role of the Chairman in each deliberation mode?

In all modes, the Chairman is auto-selected by the routing rules at STEP 1.7 in SKILL.md to synthesize the final verdict. In full mode, the Chairman produces a comprehensive document with tallies and unresolved questions. In quick mode, the Chairman generates a short recommendation with kill criteria. In duo mode, the Chairman formats the opposing positions without attempting to reconcile them into a single recommendation.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}